MiLA: Multi-view Intensive-fidelity Long-term Video Generation World Model for Autonomous Driving

arXiv 25.3 2025 59.3 method, application

TLDR

MiLA generates high-fidelity, long-duration driving videos up to one minute using coarse-to-refine and denoising modules, achieving SOTA on nuScenes.

Reasoning

The paper proposes a novel framework for long-term video generation in autonomous driving, addressing error accumulation with coarse-to-refine and denoising modules. Strengths include state-of-the-art results on nuScenes, but weaknesses are limited evaluation to a single dataset and lack of real-world deployment or interactive capabilities.

Read-first score

Read-first score 59.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 43.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Topical relevance 42%

61.4

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=dataset,experiment

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=dataset,github

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 214.

Keyword Scores

world model

generative world model

video world model

world dynamics prediction

world simulator

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Coarse-to-Re(fine) approach for stabilizing video generation and correcting distortion of dynamic objects
Temporal Progressive Denoising Scheduler
Joint Denoising and Correcting Flow modules

Methodology

MiLA is a framework for long-term video generation using a Coarse-to-Re(fine) approach to stabilize generation and correct dynamic object distortion. It incorporates a Temporal Progressive Denoising Scheduler and Joint Denoising and Correcting Flow modules. The model is trained and evaluated on the nuScenes dataset, achieving state-of-the-art performance.

Key Results

MiLA achieves state-of-the-art performance in video generation quality on the nuScenes dataset.

Resources

Paper Project